Azure Machine Learning compute of several managed CPU or GPU nodes that scales from min_instances to max_instances with each job; set the minimum to 0 and an idle scale-down time so you pay only when jobs are running.
Also called AmlCompute.
Read more: Microsoft Learn
In the Ultra Transcenders books
Each book explains Compute cluster in context, with comparison tables and the common traps.
Terms in this definition
- Azure Machine Learning
Azure platform where you train, deploy, monitor and retrain models of your own, practising MLOps through tools such as pipelines, online endpoints, model monitoring and prompt flow.
- CPU
The processor that runs a VM's workload (central processing unit); while B-series VMs run under their baseline, they accumulate CPU credits.
- GPU
A graphics processing unit, an accelerator suited to highly parallel work like graphics and AI. Azure offers them in the NV and NC VM series.
- MIN
Gives back the lowest value found in a column holding numbers, dates or text, or compares two expressions and returns the lower. For TRUE/FALSE values, which this DAX aggregation doesn't accept, turn to MINA.
- MAX
Gives back the biggest value found in a column holding numbers, dates or text, or compares two expressions and returns the bigger. For TRUE/FALSE values, which this DAX aggregation doesn't accept, turn to MAXA.
- Job
A sequence of steps executed together on one agent or runner, or on the server for agentless work. While running, each occupies one of your parallel jobs.
- Set
Secret permission in Key Vault for writing secrets; some older material refers to it as Create.
Related terms
- Batch endpoint
Endpoint type for asynchronous scoring jobs: data in storage is split into mini-batches, processed in parallel on a compute cluster, and results are saved back to storage. Choose it for large scheduled workloads, not when low latency matters.
- External storage
Lets an NCI compute cluster grow separately from its capacity by using an outside array: Dell PowerFlex or Everpure FlashArray (previously Pure Storage). Nodes set up this way give up near-sync replication and firmware updates through LCM.